AICheck365 Methodology

AICheck365 detects AI-origin clues by parsing observable provenance signals in the original file — not by running a black-box neural network. Here is what we can inspect and what we cannot guarantee.

AICheck365 can check

  • File metadata (EXIF, XMP, IPTC) — device, software, and generation tool fields
  • C2PA digital signatures and Content Credentials — embedded by OpenAI, Adobe, Microsoft, and others
  • PNG text chunks (tEXt / iTXt) — full generation parameters written by Stable Diffusion, ComfyUI, and similar tools
  • MP4 / MOV container fields — tool name, AIGC tags, SEI markers, and custom text boxes
  • Platform-specific marks — fields or patterns written by known AI platforms (Midjourney, DALL-E, Firefly, Kling, etc.)
  • Video frame watermark sampling — checking a small number of key frames for visible or invisible watermark clues when metadata yields no match
  • File-name heuristics — some platforms produce filenames with recognizable patterns
  • Container structure analysis — unusual or characteristic structures in the file format itself
  • Export / edit clues — application information written by software on re-export

AICheck365 cannot guarantee

  • Not all AI platforms embed readable provenance data in their output files; no signal does not mean human-made
  • After compression or transcoding by social platforms (WeChat, Weibo, X / Twitter, etc.), metadata is usually stripped
  • Screenshots, screen recordings, or photos of a screen carry no original file metadata
  • Platform marks have no universal standard and may change in the future
  • Signals from unknown or emerging tools may not yet be in the detection library
  • Results cannot substitute for legal appraisal or expert forensics
  • File hashes, fingerprints, and content-feature analysis are outside the scope of this tool

Evidence beats intuition. Transparency beats percentages.

Related links